Comparing methods for deriving the auditory brainstem response to continuous speech in human listeners

Several methods have recently been developed to derive the auditory brainstem response (ABR) from continuous natural speech, facilitating investigation into subcortical encoding of speech. These tools rely on deconvolution to compute the temporal response function (TRF), which models the subcortical auditory pathway as a linear system, where a nonlinearly processed stimulus is taken as the input (i.e., regressor), the electroencephalogram (EEG) data as the output, and the ABR as the impulse response deconvolved from the recorded EEG and the regressor. In this study, we analyzed EEG recordings from subjects listening to both unaltered natural speech and synthesized “peaky speech.” We compared the derived ABR TRFs using three regressors: the half-wave rectified stimulus (HWR) from Maddox and Lee (2018), the glottal pulse train (GP) from Polonenko and Maddox (2021), and the auditory nerve modeled response (ANM; Zilany et al. (2014); (2009)) used in Shan et al. (2024). Our evaluation focused on the signal-to-noise ratio, prediction accuracy, efficiency, and practicality of applying each regressor in both unaltered and peaky speech. The results indicate that the ANM regressor with peaky speech provides the best performance, with the ANM for unaltered speech and the GP regressor for peaky speech close behind, whereas the HWR regressor demonstrated relatively poorer performance. There are, thus, multiple stimulus and analysis tools that can provide high-quality subcortical TRFs, with the choices for which to use dictated by experimental needs. The findings in this study will guide future research and clinical use in selecting the most appropriate paradigm for ABR derivation from continuous, naturalistic speech.